Emotion Recognition Based on Multichannel Physiological Signals with Comprehensive Nonlinear Processing.

Area of Science:

• Physiological computing
• Machine learning for affective computing
• Biomedical signal processing

Background:

• Multichannel physiological datasets in emotion recognition are often nonlinear and difficult to process with linear methods.
• Previous linear or partially nonlinear approaches for feature reduction and classification have yielded suboptimal results.

Purpose of the Study:

• To propose a comprehensive nonlinear method for enhancing emotion recognition from physiological data.
• To address the information loss issues in traditional feature reduction techniques.

Main Methods:

• Kernel Principal Component Analysis (KPCA) with a radial basis function (RBF) kernel was used for nonlinear feature reduction, mapping data to a high-dimensional space.
• Gradient Boosting Decision Tree (GBDT), a nonlinear ensemble classifier, was employed for improved classification accuracy and feature selection.

Main Results:

• The proposed comprehensive nonlinear method demonstrated strong performance on a physiological dataset.
• Accurate classification of four distinct emotions was achieved in 29 participants with an overall accuracy of 93.42%.

Conclusions:

• The integrated nonlinear approach of KPCA and GBDT effectively handles the complexity of physiological data for emotion recognition.
• This method offers a significant advancement in accurately identifying emotions from physiological signals, outperforming traditional techniques.